Obituary for Backplane-Based Systems and Connectors
Demand Falling for Backplane-Based Systems
Let’s face reality, folks: The demand for backplane-based systems is falling like a refrigerator down an elevator shaft (as is the demand for copper-based backplane connectors). There are several logical reasons for this precipitous decline, and a set of laws that explain why this is happening. But first, we have to look at the history of computer systems to understand the source of the necrosis.
From the 1940s through about 2005, all computer systems were processor-bound: The interconnects could feed more data than the CPU could process. We used shared multi-drop buses pushing single-ended signals down a copper backplane trace through pins on copper connectors. This worked for a long time, since processor speeds increased ever so slowly. We went through numerous trivial incremental improvements in bus drivers, from TTL to LS (low-power Schottky), and finally to BTL (Backplane Transceiver Logic). But this technology model fell apart in the 2000s when we hit the signal integrity wall at about 200MHz.
Around 2005, we hit a tipping point. The tremendous advances in the clock frequencies of modern CPUs made all computer systems I/O-bound: The processor could process more data than the interconnects could deliver. That pushed us into LVDS (low-voltage differential signaling) and high-speed serial fabric point-to-point links on copper. Today, we are still horribly I/O-bound in all computers, even when pushing data at 1G, 3G, 5G, 8G, and 10G rates into the CPU (especially in multi-core machines). To increase the aggregate throughput of the backplane and feed the processors, we started using multiple fabric links (X2, X4, X8, and X16) between the boards. This methodology sucked up copper pins in the connectors at an alarming rate. With today’s high-speed copper connectors, we now use hundreds of pins per slot for a 6U board, just to feed the CPUs with data. And we are still I/O-bound.
The transceiver guys are working their magic with new techniques (equalization; new, more complex signaling protocols; etc.) and the backplane designers are using better FR4-type materials and trace impedance control processes to keep the copper party going for a while longer. But at some point we will hit the signal integrity wall again, encounter another tipping point, and suffer the same disaster experienced by buses on copper-based backplanes using copper-based connectors.
The Immutable Laws You Never Suspected
Where is that wall? Probably somewhere between 14 and 20GHz, the new tipping point. That event will then invoke the natural laws governing copper backplanes and copper connectors. The curves these laws create are not perfectly linear Euclidean geometry vectors. They are not perfectly curvilinear Riemann geometry functions either. They look more like unequal negative step-functions of maximum intelligible signal-distance versus signal-frequency graphs.
Law 1: [ when (f+P)= +, then (s/n)= -]
As the frequency (f) increases and signaling protocol (P) becomes more complex, the signal-to-noise ratio (s/n) declines. The higher the frequency and the more complex the signaling protocol being pushed along copper traces and through copper connector contacts, the more the signal looks like noise.
Hephaestus (the Greek god of metallurgy) is powerless to solve this problem. The Cuprozites (the connector people, who worship molten calves made of copper and believe in weak molecular bonding) have been riding the lethargic incremental curve of the Siliconites (the semiconductor people, who worship idols made of silicon and believe in strong crystalline bonding). This has been a parasitic relationship for decades. Even Crystalus, the god of silicon, has no clue how to push anything but noise through copper connectors at 15G-20G and down a copper trace on a backplane. This is going to be a hard fight with a very short stick for the Cuprozites in the connector business. Crystalus is a much more powerful god than Hephaestus, and has blessed his loyal followers with a divine solution — optical links.
Law 2: (2Bf=d/2)
Every time the frequency of the signals going through a copper connector and along copper backplane traces doubles (2Bf), the distance those signals can run declines by 50% (d/2).
This was particularly true when we used buses and single-ended signals as interconnects. This law remains true for serial-differential interconnects, but the distance the signals can run on copper backplane traces haven’t declined by as much as 50% over the frequencies we have experienced so far. As we move beyond 10G, the distance the signals can run on the copper traces declines by more than 50%. At some point (15G-20G?), Law 1 will bite the connector guys in a tender place, infecting them with the froth of feral hydrophobic signal integrity horrors, for which there is only one known cure: Optical links.
Law 3: (2Nf=B/2)
Every time the network link frequency doubles (2Nf), the demand for copper-based backplanes and their associated connectors declines by 50% (B/2).
Backplane-based systems are centralized systems. They traditionally offered higher data transfer speeds on the backplane, between the boards, than the network links could offer between boxes. Besides providing performance advantages over network connections, backplane-based systems also have another major benefit: Modularity. Modularity, in turn, gives us two major capabilities that many critical systems desire – maintainability and upgradeability. Additionally, the boards in centralized backplane systems can be cooled more efficiently, whether with air, conduction cooling, or liquid. However, as the network link bandwidth on a fiber cable increases past the bandwidth of the copper backplane trace, the centralized system of boards in a rack can be easily broken apart into a distributed system of motherboards and small form factor boxes, all connected with the faster network connections. At that point, copper will become a hazardous substance and the demand for copper backplane connectors will be slightly larger than the market for kosher ham. This law is coming into effect now as the data centers migrate from copper twisted pair and coax to 40G and 100G optical network links. The data center server boards were migrating to copper coax cables in the short term, but are now moving to AOCs (active optical cables) between the server blades, to increase the bandwidth between the boards in the rack to something equal to or greater than the bandwidth of the network links.
Law 4: (if Nf > Bf, then 2Nf = B/2)
To maintain the modularity benefits of backplane-based centralized systems, the bandwidth of each data link on the backplane (Bf), between the boards, must be greater than the bandwidth of the network link cable (Nf).
If the network link bandwidth in a distributed system architecture (Nf) is greater than the backplane link bandwidth between the boards in the centralized system architecture (Bf), then backplanes lose their primary performance advantage and Law 3 is invoked. The network performance advantage would then trump the modularity benefits of maintainability and upgradeability in the copper-based backplane system, and those racks of boards will be split apart into distributed systems (B/2). We saw this happen in the 1970s, when Datapoint started connecting small minicomputers in a distributed system with early Ethernet connections. That innovation destroyed the centralized-architecture, copper-backplane-based mainframe computer market. We are seeing this law ensue in data centers today, as they move to optical network links.
A Little History of These Laws’ Effects
You probably doubt these laws at this point, for some esoteric, illogical, and ingrained reason, so let’s look at some industries where they have taken over. The industrial control and machine control markets used backplane-based systems for many years (STD, MB1 and MB2, VME, cPCI, etc.). In the past 10 years, most of these apps have moved to SFF (small form factor) and motherboards. Industrial systems are sequencers that emulate relay logic, and the present processor and I/O technologies are overkill for their needs. The industrial market is cost-driven now, not performance-driven.
The medical market (CAT scan, CT scan, MRI, etc.) used backplane-based systems for many years. They have transitioned to motherboards, since their I/O bandwidth requirements are also satiated. They have no need for the benefits of backplanes any longer. The medical equipment markets are also cost-driven now, not performance-driven.
Two remaining markets desire the benefits of modularity: military systems and telecom systems. The military applications are driving the need for higher I/O performance, more so than any other market segment. These needs show up in RADAR, SONAR, electronic warfare, signal intelligence, and communications intelligence systems. We are seeing more and more optical links going into UAVs, ships, and aircraft for the weight, speed, and noise immunity benefits.
Telecoms have needs for higher performance backplane-based systems, but they are not willing to pay much for them. Many companies have crashed upon the shores of Telecom Island trying to make a profit selling commodity products to commodity telecom buyers. Telecoms are accustomed to rooms full of racks full of computer boards, but the revenues for most telecom service providers are growing slowly or falling. They buy the cheapest computer cycles they can find and live with the bandwidth limitations by buying more of the commodity boards. They use optical links for backhaul and long-haul communications only, where they can aggregate a lot of data on a single link.
The biggest market demanding bandwidth is the data centers. As stated before, they have been moving to optical links for years. They use 3U size pizza-box server boards, and get their upgradeability and maintainability benefits at the shelf level (instead of the board level). So, they have already broken apart those centralized blade server backplane-based racks into a distributed system.
Better Do Some Analysis
We have established that optical links will replace copper links on backplanes in centralized systems in the near future, in several market segments. Optical is a destructive innovation for copper-based backplanes and connectors. So now, it’s time to explore that with some analysis. I will do this analysis for you for a large sum of money, or you can do it yourself for about $50 worth of books.
First, you need to understand “tipping points,” how they develop, and how they operate. Go buy The Tipping Point by Malcolm Gladwell and read it. This will be a foundation for the other books in this list.
Second, you need to understand destructive innovations and how they affect you and your company. Go buy The Dark Side of Innovation by Ankush Chopra and read it. This will give you a complete understanding of your options and strategy when innovations show up that can destroy your product line and your finances.
Third, you need to be able to predict the rate of diffusion of destructive technologies. Go buy Diffusion of Innovations (fifth edition) by Everett Rogers and understand the “Rogers Model.”
Fourth, take what you have learned in these books and apply it to your company. I can see numerous potential hybrid optical-copper connector possibilities that will be very beneficial to the remaining backplane-based computer markets. I can also see potential for genderless and blind-mate optical connectors that can handle severe shock, vibration, and temperature extremes. There are amazing opportunities for optical-based connectors in the future, if you abandon Hephaestus and start worshiping Crystalus.
Wrapping This Up
It’s pretty clear, at this point, that copper backplanes and copper backplane connectors are just brothels housing promiscuous electrons. Photons have much higher morals and can eliminate many of the transgressions that copper backplanes and copper connectors commit. Hopefully, this article will stir some mental effort in the cube farms of the backplane connector vendors, and they will see the light.
By Ray Alderman, Chairman of the Board of Directors, VITA
Ray Alderman is chairman of the Board of Directors at VITA. He was in military intelligence in the Vietnam War; started in the mainframe computer business with Burroughs; was a partner in several computer start-ups; president of PEP Modular Computers; and enjoys irritating connector vendors as a hobby. Please feel free to harass him at [email protected].